Github-Mirror/seaweedfs
1
0 Fork 0
mirror of https://github.com/seaweedfs/seaweedfs.git synced 2026-10-31 17:30:42 +01:00
Code Issues Packages Projects Releases Wiki Activity

parquet_pushdown: distinguish "no score" from Score == 0 in ScoredRowRef

Browse Source 
A bare float32 score conflates "this row has no score, it came from
a scalar predicate" with "this row's vector similarity score is
0.0", which is a legal value: zero L2 distance and zero dot product
both arise in real queries. Mixing scored and unscored results in
one response (e.g. hybrid scalar+vector) becomes ambiguous.

Add has_score (proto field 2, sandwiched between Ref and Score) and
the corresponding HasScore field in the design's Go sketch. Update
the prose paragraph to call out the new flag's purpose.
This commit is contained in:
Chris Lu committed 2026-04-25 15:09:15 -07:00
1 parent 6c826e710a
commit 8fe0b68b7c
3 files changed
+35 -13
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

weed/pb/parquet_pushdown.proto
+6 -1
View File
@@ -220,7 +220,12 @@ message RowRef {
message ScoredRowRef {
RowRef ref = 1;
float score = 2;
// has_score distinguishes "no score available" from "score is
// 0.0", which is a legal vector-similarity value (zero L2
// distance, zero dot product). Non-vector queries leave
// has_score false; vector queries always set it true.
bool has_score = 2;
float score = 3;
}
message PushdownStats {
weed/pb/parquet_pushdown_pb/parquet_pushdown.pb.go
+19 -6
View File
@@ -1189,9 +1189,14 @@ func (x *RowRef) GetFilePosition() int64 {
}
type ScoredRowRef struct {
state protoimpl.MessageState `protogen:"open.v1"`
Ref *RowRef `protobuf:"bytes,1,opt,name=ref,proto3" json:"ref,omitempty"`
Score float32 `protobuf:"fixed32,2,opt,name=score,proto3" json:"score,omitempty"`
state protoimpl.MessageState `protogen:"open.v1"`
Ref *RowRef `protobuf:"bytes,1,opt,name=ref,proto3" json:"ref,omitempty"`
// has_score distinguishes "no score available" from "score is
// 0.0", which is a legal vector-similarity value (zero L2
// distance, zero dot product). Non-vector queries leave
// has_score false; vector queries always set it true.
HasScore bool `protobuf:"varint,2,opt,name=has_score,json=hasScore,proto3" json:"has_score,omitempty"`
Score float32 `protobuf:"fixed32,3,opt,name=score,proto3" json:"score,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
@@ -1233,6 +1238,13 @@ func (x *ScoredRowRef) GetRef() *RowRef {
return nil
}
func (x *ScoredRowRef) GetHasScore() bool {
if x != nil {
return x.HasScore
}
return false
}
func (x *ScoredRowRef) GetScore() float32 {
if x != nil {
return x.Score
@@ -1444,10 +1456,11 @@ const file_parquet_pushdown_proto_rawDesc = "" +
"\x06RowRef\x12\x12\n" +
"\x04file\x18\x01 \x01(\tR\x04file\x12\x1b\n" +
"\trow_group\x18\x02 \x01(\x05R\browGroup\x12#\n" +
"\rfile_position\x18\x03 \x01(\x03R\ffilePosition\"S\n" +
"\rfile_position\x18\x03 \x01(\x03R\ffilePosition\"p\n" +
"\fScoredRowRef\x12-\n" +
"\x03ref\x18\x01 \x01(\v2\x1b.parquet_pushdown_pb.RowRefR\x03ref\x12\x14\n" +
"\x05score\x18\x02 \x01(\x02R\x05score\"\x81\x03\n" +
"\x03ref\x18\x01 \x01(\v2\x1b.parquet_pushdown_pb.RowRefR\x03ref\x12\x1b\n" +
"\thas_score\x18\x02 \x01(\bR\bhasScore\x12\x14\n" +
"\x05score\x18\x03 \x01(\x02R\x05score\"\x81\x03\n" +
"\rPushdownStats\x12\x1d\n" +
"\n" +
"trust_mode\x18\x01 \x01(\tR\ttrustMode\x12,\n" +